From 30e57f6e6346ced3a643c2cb5497862968d05f64 Mon Sep 17 00:00:00 2001 From: Imran Khan Date: Thu, 1 Aug 2024 14:46:58 +1000 Subject: [PATCH 1/2] kernfs,memcg: Add helpers to get memcg info Add kernfs based helpers to extract memcg related information, like number of active and inactive memcgroups, page cache pages pinning memcgroups etc. Orabug: 37322867 Signed-off-by: Imran Khan --- drgn_tools/kernfs_memcg.py | 251 +++++++++++++++++++++++++++++++++++++ 1 file changed, 251 insertions(+) create mode 100644 drgn_tools/kernfs_memcg.py diff --git a/drgn_tools/kernfs_memcg.py b/drgn_tools/kernfs_memcg.py new file mode 100644 index 00000000..952fee9e --- /dev/null +++ b/drgn_tools/kernfs_memcg.py @@ -0,0 +1,251 @@ +# Copyright (c) 2024, Oracle and/or its affiliates. +# Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/ +""" +Kernfs_memcg +-------------- + +The ``drgn.helpers.linux.kernfs_memcg`` module provides helpers for working with the +Linux memcg subsystem. +""" +import argparse +from typing import Iterator + +from drgn import cast +from drgn import container_of +from drgn import FaultError +from drgn import Object +from drgn import Program +from drgn.helpers.common.format import decode_enum_type_flags +from drgn.helpers.linux import cgroup_path +from drgn.helpers.linux import css_for_each_descendant_pre +from drgn.helpers.linux import find_slab_cache +from drgn.helpers.linux import for_each_page +from drgn.helpers.linux import inode_path +from drgn.helpers.linux import kernfs_path +from drgn.helpers.linux import PageSlab +from drgn.helpers.linux import slab_cache_for_each_allocated_object +from drgn.helpers.linux import slab_cache_is_merged + +from drgn_tools.corelens import CorelensModule +from drgn_tools.dentry import dentry_path_any_mount + + +_KERNFS_TYPE_MASK = 0xF +# cgroup subsystem id for memory cgroup, from kernel/cgroup/cgroup.c +_MEMORY_CGRP_ID = 4 + + +def decode_css_flags(css: Object) -> str: + """ + Get a human-readable representation of cgroup_subsys_state.flags + + :param css: ``struct cgroup_subsys_state *`` + """ + CSS_DYING = css.prog_["CSS_DYING"] + flags = css.flags.value_() + if not flags: + # There is no dedicated flag value to indicate a zombie cgroup. + # A css.flags value of 0 indicates that cgroup destruction is + # complete but cgroup object has not been fully freed because + # of being pinned by some other object + return "ZOMBIE" + + return decode_enum_type_flags(flags, CSS_DYING.type_, False) + + +def for_each_kernfs_node(prog: Program) -> Iterator[Object]: + """ + Iterate over all kernfs_node objects in the system. + + :returns: Iterator of ``struct kernfs_node *`` objects. + """ + kernfs_node_cache = find_slab_cache(prog, "kernfs_node_cache") + if not slab_cache_is_merged(kernfs_node_cache): + for kn in slab_cache_for_each_allocated_object( + kernfs_node_cache, "struct kernfs_node" + ): + yield kn + else: + print( + "Can't iterate through kernfs_node_cache, because its merged with other slab-cache(s)" + ) + + +def dump_memcgroup_hierarchy(prog: Program) -> None: + """ + Dump hierarchy of active mem cgroups. + """ + cgroup_subsys = prog["cgroup_subsys"][_MEMORY_CGRP_ID] + css = cgroup_subsys.root.cgrp.self.address_of_() + print(f"dumping: {cgroup_subsys.name.string_().decode()} hierarchy") + for pos in css_for_each_descendant_pre(css): + cgroup_state = decode_css_flags(pos) + print( + f"path: {cgroup_path(pos.cgroup).decode()} state: {cgroup_state}" + ) + + +def kernfs_node_of_cgroup(kn: Object) -> bool: + """ + Check if a kernfs_node object represents a cgroup object. + + :param kn: ``struct kernfs_node *`` + :returns: True if kernfs_node object represents a cgroup object, + False otherwise. + """ + if (kn.flags.value_() & _KERNFS_TYPE_MASK) == kn.prog_.constant( + "KERNFS_DIR" + ).value_(): + try: + cgrp = cast("struct cgroup *", kn.priv) + return cgrp.kn == kn + except FaultError: + return False + else: + return False + + +def kernfs_node_of_memcgroup(kn: Object) -> bool: + """ + Check if a kernfs_node object represents a mem cgroup object. + + :param kn: ``struct kernfs_node *`` + :returns: True if kernfs_node object represents a mem cgroup object, + False otherwise. + """ + if kernfs_node_of_cgroup(kn): + prog = kn.prog_ + cgrp = cast("struct cgroup *", kn.priv) + return prog["cgroup_subsys"][_MEMORY_CGRP_ID].root == cgrp.root + else: + return False + + +def dump_memcg_kernfs_nodes(prog: Program) -> None: + """ + List all kernfs_node objects that represent a mem cgroup. + """ + count = 0 + for kn in for_each_kernfs_node(prog): + if kernfs_node_of_memcgroup(kn): + count = count + 1 + path = kernfs_path(kn).decode() + print("kernfs_node: ", hex(kn.value_()), " ", path) + + print("Total number of memcg kernfs_node objects: ", count) + + +def get_num_active_mem_cgroups(prog: Program) -> int: + """ + Get number of active mem cgroups. + """ + mem_cgroup_subsys = prog["cgroup_subsys"][_MEMORY_CGRP_ID] + # add 1 to number of active memcgroups to account for root memcgroup + return mem_cgroup_subsys.root.cgrp.nr_descendants.value_() + 1 + + +def get_num_dying_mem_cgroups(prog: Program) -> int: + """ + Get number of inactive or dying mem cgroups. + """ + mem_cgroup_subsys = prog["cgroup_subsys"][_MEMORY_CGRP_ID] + return mem_cgroup_subsys.root.cgrp.nr_dying_descendants.value_() + + +def get_num_mem_cgroups(prog: Program) -> None: + active_mem_cgroups = get_num_active_mem_cgroups(prog) + dying_mem_cgroups = get_num_dying_mem_cgroups(prog) + print( + f"There are {active_mem_cgroups} active and {dying_mem_cgroups} dying memcgroups \n" + ) + + +# By default (max_pages == 0) we scan all pages, +# that have memcg ref but if max_pages is specified +# then we bail out after getting those many pages +# or after scanning all pages , whichever happens first. +def dump_page_cache_pages_pinning_cgroups(prog: Program, max_pages: int = 0): + """ + Dump page-cache pages that have reference to a mem-cgroup. + The ouput also contains information such as the cgroup that is pinned, its flags + (to indicate current state of cgroup) and file cached by this page. + + :params: max_pages: specify how many pages to find. By default first 10000 such + pages are listed. Use 0 to list all such pages. + + """ + mem_cgroup_root = prog["cgroup_subsys"][_MEMORY_CGRP_ID].root + total_count = 0 + found_count = 0 + fault_count = 0 + for page in for_each_page(prog): + total_count = total_count + 1 + try: + # Ignore slab pages + if PageSlab(page): + continue + # Ignore non page-cache pages + if not page.mapping: + continue + try: + mem_cgroup = page.mem_cgroup + except AttributeError: + mem_cgroup = page.memcg_data + + if not mem_cgroup.value_() or mem_cgroup.value_() & 3: + continue + cgroup_subsys_state = cast( + "struct cgroup_subsys_state *", mem_cgroup + ) + if cgroup_subsys_state.cgroup.root == mem_cgroup_root: + found_count = found_count + 1 + cgrp = cgroup_subsys_state.cgroup + address_space = page.mapping + inode = address_space.host + if inode_path(inode) is None: + continue + dentry = container_of( + inode.i_dentry.first, "struct dentry", "d_u.d_alias" + ) + path = dentry_path_any_mount(dentry).decode() + cgroup_state = decode_css_flags(cgrp.self.address_of_()) + print( + f"page: 0x{page.value_():x} cgroup: {cgroup_path(cgrp).decode()} state: {cgroup_state} path: {path}\n" + ) + if max_pages and found_count == max_pages: + break + except FaultError: + fault_count = fault_count + 1 + continue + + print( + f"Scanned {total_count} pages, found {found_count} pages with memory cgroup refs, found {fault_count} faults." + ) + + +class NumMemCgroups(CorelensModule): + """Print number of active and dying memcgroups""" + + name = "num-memcgroups" + + def run(self, prog: Program, args: argparse.Namespace) -> None: + get_num_mem_cgroups(prog) + + +class PagesPinningMemcgroups(CorelensModule): + """Print information related to pages, that are pinning memcgroup(s)""" + + name = "pages-pinning-memcg" + run_when = "never" + + def add_args(self, parser: argparse.ArgumentParser) -> None: + parser.add_argument( + "--max", + "-m", + type=int, + default=0, + help="Maximum number of pages to show. By default(0) all such pages are shown.", + ) + + def run(self, prog: Program, args: argparse.Namespace) -> None: + dump_page_cache_pages_pinning_cgroups(prog, max_pages=args.max) From 7636da04a90a72380453b5dbcb732c1d19066c82 Mon Sep 17 00:00:00 2001 From: Imran Khan Date: Tue, 22 Oct 2024 21:41:21 +1100 Subject: [PATCH 2/2] kernfs,memcg: Add test case for memcg helpers. Orabug: 37322867 Signed-off-by: Imran Khan --- tests/test_kernfs_memcg.py | 33 +++++++++++++++++++++++++++++++++ 1 file changed, 33 insertions(+) create mode 100644 tests/test_kernfs_memcg.py diff --git a/tests/test_kernfs_memcg.py b/tests/test_kernfs_memcg.py new file mode 100644 index 00000000..d4924f99 --- /dev/null +++ b/tests/test_kernfs_memcg.py @@ -0,0 +1,33 @@ +# Copyright (c) 2024, Oracle and/or its affiliates. +# Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/ +import drgn + +from drgn_tools import kernfs_memcg as kernfs_memcg + + +def test_dump_page_cache_pages_pinning_cgroups(prog: drgn.Program) -> None: + kernfs_memcg.dump_page_cache_pages_pinning_cgroups(prog, 10) + + +def test_dump_memcgroup_hierarchy(prog: drgn.Program) -> None: + kernfs_memcg.dump_memcgroup_hierarchy(prog) + + +def test_kernfs_node_of_memcgroup(prog: drgn.Program) -> None: + count = 0 + for kn in kernfs_memcg.for_each_kernfs_node(prog): + if kernfs_memcg.kernfs_node_of_memcgroup(kn): + count = count + 1 + if count >= 5: + print("Found 5 memcgroup, kernfs_node objects.") + break + + +def test_get_num_active_mem_cgroups(prog: drgn.Program) -> None: + count = kernfs_memcg.get_num_active_mem_cgroups(prog) + print(f"number of active memcgroups: {count}\n") + + +def test_get_num_dying_mem_cgroups(prog: drgn.Program) -> None: + count = kernfs_memcg.get_num_dying_mem_cgroups(prog) + print(f"number of dying memcgroups: {count}\n")